Research On Feature Selection Method Of Software Defect Prediction

Software defects will inevitably occur when software projects are developed and studied.Therefore,the timely detection and elimination of the defects of the related software developed has become the key work that software project engineering development needs to focus on.With the continuous development of global economic science and technology,in the current era of knowledge economy,software products are full of people's daily life,at the same time,the role of software products in the field of social production and life continues to improve.Accordingly,the quality of software products and systems is becoming more and more concerned.As the key content of software engineering,software defect prediction technology is based on relevant experience data,with the help of machine The method of learning can help software developers Sand users to find relevant software defects in time,and then effectively save the resources needed for software development and improve the efficiency of software development to ensure the quality of products.After years of development,software prediction technology has made great research progress,but there are still shortcomings,such as the classification of prediction models is not accurate,the applicability and pertinence of prediction method selection are not strong,which also limits the application of software prediction technology in related industries to some extent,and also increases the hidden dangers of related software systems and products.Feature selection is the process of optimizing the system specific indicators by selecting N valid features from the existing M features and then reducing the dimension of the data set.feature selection is not only an important data preprocessing technique in traditional pattern recognition,but also an effective means to improve learning algorithms.it mainly includes two types of algorithms,Filter and Wrapper.the difference between the two is mainly reflected in the evaluation of N subset of features.The Filter algorithm is mainly based on the mathematical characteristics of the data set,and the Wrapper algorithm is based on the learning algorithm Both algorithms have their own advantages.A large number of research practices show that by applying the relevant feature algorithms to the prediction of software defect models,we can find and make up for the defects in the software and improve the quality of software products.Based on this,based on reading a large number of domestic and foreign literature on software defect prediction and feature selection methods,combined with their own professional knowledge,this paper expounds and analyzes the related concepts of software defect and feature selection,and puts forward two feature selection algorithms based on the concept of software defect and feature selection:A feature selection based on mutual information improvement,which belongs to the Filter algorithm.considering that the evaluation criteria are affected by the feature subset,a feature subset evaluation function is proposed.the salient feature of this evaluation function is the introduction of feature sub-ons with nonlinear factors,the improvement of the existing mutual information are improved,and the accuracy of software defect prediction model classification is improved by experimental research.Experiments show that the feature selection algorithm based on mutual information has obvious advantages in improving the classification performance of software defect prediction model.The other is based on genetic support The validity of the algorithm is further verified.The results show that the two feature selection algorithms proposed in this paper have obvious effect on improving the classification accuracy of software defect prediction model and eliminating the redundancy of software defect data set.